Drug–Drug Interaction (DDI) relation extraction based on deep learning is a critical task in drug development and pharmacovigilance. DDI affects treatment planning, monitoring effects of medicine, and patient safety, making it essential for researchers and doctors. Existing DDI datasets have limited positive instances, challenging deep learning models to extract sufficient features from text. To address this, researchers use feature supplementation methods to collect features from various data types. This review summarizes the general process of DDI relation extraction using deep learning, discusses feature supplementation methods, and reviews state-of-the-art literature. It compares feature supplementation methods and provides suggestions for future research. The purpose is to help researchers understand feature complementation methods and design custom DDI relation extraction methods. The review covers preprocessing, word embedding, feature supplementation, and network frameworks. It discusses internal and external feature supplementation methods, including position and POS embeddings, dependency and SDP-based methods, network combination, and special preprocessing. External methods include text data and molecular structure information, as well as biomedical embeddings. Deep learning frameworks such as CNN, RNN, GCN, and simplified networks are reviewed. Performance evaluation shows that methods like DeepCNN1 and DeepCNN2 achieve high performance. Challenges include data imbalance, limited feature information, and the need for more efficient models. Future research should focus on improving BERT models and utilizing external data for better DDI relation extraction.Drug–Drug Interaction (DDI) relation extraction based on deep learning is a critical task in drug development and pharmacovigilance. DDI affects treatment planning, monitoring effects of medicine, and patient safety, making it essential for researchers and doctors. Existing DDI datasets have limited positive instances, challenging deep learning models to extract sufficient features from text. To address this, researchers use feature supplementation methods to collect features from various data types. This review summarizes the general process of DDI relation extraction using deep learning, discusses feature supplementation methods, and reviews state-of-the-art literature. It compares feature supplementation methods and provides suggestions for future research. The purpose is to help researchers understand feature complementation methods and design custom DDI relation extraction methods. The review covers preprocessing, word embedding, feature supplementation, and network frameworks. It discusses internal and external feature supplementation methods, including position and POS embeddings, dependency and SDP-based methods, network combination, and special preprocessing. External methods include text data and molecular structure information, as well as biomedical embeddings. Deep learning frameworks such as CNN, RNN, GCN, and simplified networks are reviewed. Performance evaluation shows that methods like DeepCNN1 and DeepCNN2 achieve high performance. Challenges include data imbalance, limited feature information, and the need for more efficient models. Future research should focus on improving BERT models and utilizing external data for better DDI relation extraction.